Professional Documents
Culture Documents
Cognitive Neuroscience
Abstract
■ Noninvasive brain stimulation (NIBS) techniques, such as and in connected remote sites of the network, which in conse-
transcranial magnetic stimulation or transcranial direct and al- quence affects the cognitive function of interest and eventually
ternating current stimulation, are advocated as measures to en- results in a change of the behavioral measure. Importantly, ev-
able causal inference in cognitive neuroscience experiments. ery link in this causal chain of effects can be confounded by
Transcending the limitations of purely correlative neuroimaging several factors that have to be experimentally eliminated or con-
measures and experimental sensory stimulation, they allow to trolled to attribute the observed results to their assumed cause.
experimentally manipulate brain activity and study its conse- This is complicated by the fact that many of the mediating and
quences for perception, cognition, and eventually, behavior. confounding variables are not directly observable and dose–
Although this is true in principle, particular caution is advised response relationships are often nonlinear. We will walk the
when interpreting brain stimulation experiments in a causal reader through the chain of causation for a generic cognitive
manner. Research hypotheses are often oversimplified, disre- neuroscience NIBS study, discuss possible confounds, and ad-
garding the underlying (implicitly assumed) complex chain of vise appropriate control conditions. If crucial assumptions are
causation, namely, that the stimulation technique has to gener- explicitly tested (where possible) and confounds are experi-
ate an electric field in the brain tissue, which then evokes or mentally well controlled, NIBS can indeed reveal cause–effect
modulates neuronal activity both locally in the target region relationships in cognitive neuroscience studies. ■
INTRODUCTION
were discovered as methods for subthreshold modula-
Noninvasive brain stimulation (NIBS) techniques, such as tion of neuronal activity and thus of cognitive function
TMS or transcranial direct and alternating current stimu- ( Vosskuhl, Strüber, & Herrmann, 2018; Yavari, Jamil,
lation (TDCS/TACS), allow to experimentally manipulate Mosayebi Samani, Vidor, & Nitsche, 2018). Although cog-
neuronal activity in the healthy human brain in a tempo- nitive neuroscience studies using these NIBS techniques
rally and spatially specific manner, thereby overcoming often claim to test the “causal relevance” of a specific
the merely correlative nature of electrophysiological brain region or neuronal activity pattern for a specific
and neuroimaging techniques (Bergmann, Karabanov, cognitive function or behavior, the underlying cause–
Hartwigsen, Thielscher, & Siebner, 2016). Their ability effect relationships are rarely made explicit. However,
to bypass sensory input channels and directly affect brain to justify such causal inference, the theoretically assumed
activity makes them unparalleled tools for studying chain of causation, leading from the applied stimulation
cause–effect relationships between neuronal activity to the observed behavioral change, has to hold for a con-
and cognitive function. Shortly after its invention crete empirical experiment. Crucially, every single cause–
(Barker, Jalinous, & Freeston, 1985), TMS was already effect link in this causal chain can be interrupted or
demonstrated to be capable of suppressing visual percep- confounded by several factors, which are best eliminated
tion (Amassian et al., 1989), and by now, the “disruptive” or controlled experimentally to attribute the observed
or “interfering” effects of TMS have a long-standing tradi- results to their assumed cause. We will start by briefly
tion in cognitive neuroscience, following the so-called outlining the core element of this article: a simplified
“virtual lesion” approach (Pascual-Leone, Walsh, & five-step chain of causation for cognitive neuroscience
Rothwell, 2000; Walsh & Cowey, 2000). Later, TDCS NIBS studies and its principal confounders. We will then
(Nitsche & Paulus, 2000) and TACS (Antal et al., 2008) introduce general experimental approaches using NIBS
and discuss the concept of causal inference for the case
of experimental NIBS studies in cognitive neuroscience
1
Leibniz Institute for Resilience Research, Mainz, Germany, before we walk the reader step-by-step through the
2 five-step chain of causation. Afterward, we will discuss
Max Planck Institute for Human Cognitive and Brain Sciences,
Leipzig, Germany potential confounders in more detail and review the
© 2020 by the Massachusetts Institute of Technology. Published under Journal of Cognitive Neuroscience X:Y, pp. 1–29
a Creative Commons Attribution 4.0 International (CC BY 4.0) license. https://doi.org/10.1162/jocn_a_01591
available experimental control conditions to counteract Figure 1B, we provide a more elaborated causal diagram,
them before we conclude by providing 12 general recom- but for the sake of comprehensibility, it is still simplified
mendations for designing valid NIBS studies. and reduced to the key variables (some of them summa-
Please note that, in the context of this article, causal in- rizing multiple smaller ones). The red elements connected
ference simply means “inferring causality” or “inferring by red arrows indicate the core chain of causation leading
that one variable is the cause of another” (Scheines, from application of NIBS to an observable behavioral ef-
2005), an inference that may either be based on the con- fect. The yellow box and arrows indicate the causal route
trolled randomized experiment or, under certain condi- by which task demands and the current brain state drive
tions, on observational data alone, when using the causal local and network activity and thereby the respective cog-
inference framework developed by Judea Pearl (Pearl & nitive function and behavioral responses. This causal path
Mackenzie, 2018; Pearl, 2010) and others. A more general is not only a source of confounding but is relevant for task
introduction into the latter and its implications for neuro- performance in the absence of NIBS and the main drive for
imaging studies is outside the scope of this article. While the cognitive function of interest. In fact, NIBS-induced
adopting elements from this particular causal inference brain activity per se is insufficient to cause more complex
framework, this review remains largely focused on the clas- cognitive functions and only modulates ongoing task-
sical experimentalist’s framework of inferring causation via related neuronal activity. The black arrows indicate addi-
experimental manipulation. We primarily aim to raise tional causal relationships that result from or eventually af-
awareness for the underlying (often implicitly assumed) fect elements of the main causal chain of interest and may
chain of causation in NIBS studies, their potential con- thereby produce confounding via central (blue boxes) and
founds, and respective experimental control measures, peripheral (green boxes) off-target stimulation. We segre-
encouraging the conduction of well-planned and well- gated the core chain of causation from NIBS application to
controlled NIBS experiments that actually justify causal behavior into five cause–effect pairs as described below.
inference, that is, the conclusion of cause–effect relation- Arrow 1: The application of NIBS produces an electric
ships between neuronal activity and cognitive function. field (E-field) in the brain tissue, either via transcranial
electro-magneto-electric induction (TMS) or the direct
administration of weak transcranial currents (transcranial
The Chain of Causation in NIBS Studies
current stimulation [TCS]).
A chain of causation (or causal chain) refers to an unin- Arrow 2: The E-field then interacts with the neurons’
terrupted concatenation of cause–effect pairs, leading membrane potential to immediately (online) evoke neu-
from an initial cause of interest via a number of mediating ronal firing (TMS) or modulate the threshold for sponta-
variables to an eventual effect of interest. Given that most neous firing (TCS) locally in the targeted brain region,
effects have many causes and themselves cause many ef- activating specific intracortical circuit motifs and asso-
fects, such a chain represents only one specific path ciated neuronal signatures ( Womelsdorf, Valiante, Sahin,
through an entire causal diagram. A causal diagram can Miller, & Tiesinga, 2014). In the case of offline TMS/TCS
be formalized as variables connected with arrows that in- protocols, it additionally triggers processes of synaptic
dicate causation (A ➔ B) instead of mere association (A, plasticity.
B), with the left and right variables representing cause Arrow 3: If driving output neurons of the targeted net-
and effect for a particular cause–effect pair (Pearl & work node to suprathreshold levels, local neuronal activ-
Mackenzie, 2018). Importantly, there is no ultimate cause ity transsynaptically spreads to other connected brain
or effect, and the partial chain to be considered depends regions of the targeted network via intercortical axonal
entirely on the research question at hand. Once the hy- projections, activating large-scale and remote circuit mo-
pothesis has defined cause and effect of interest, any in- tifs as well as remote changes in synaptic strength.
termediate element within a causal path connecting them Arrow 4: The immediate (online) or subsequent (off-
is referred to as “mediator.” In contrast, elements that are line) effects on local and large-scale circuit motifs can dis-
associated with the cause of interest (i.e., causing it or turb or facilitate the specific task-relevant neuronal
merely covarying with it but not caused by it) and that computations mediating the cognitive function of inter-
influence the effect of interest are considered con- est, which is reflected in either respective changes of
founders, because their influence on the effect of interest the outcome or the completion time of these processes.
is mixed with that of the cause of interest, preventing the Arrow 5: With regard to the motor responses exerted
straightforward attribution of causal influence to the lat- in the context of a specific behavioral task, these altered
ter and therefore the identification of an unambiguous cognitive processes eventually result in changed error
causal path through the causal diagram. rates or RTs.
The somewhat naive level at which hypotheses are often
phrased in NIBS studies is depicted in Figure 1A. The stim-
Confounding in NIBS Studies
ulation is expected to affect a single circumscribed brain
region, which has an effect on behavior, because the brain Most cognitive neuroscience studies investigate the neuro-
region is causally relevant for producing that behavior. In nal implementation of a cognitive function and therefore
hypothesize that neuronal activity in a certain brain region is relationships. Depending on the expertise and educational
causal to that cognitive function of interest. NIBS is merely background of the researcher, some of these links are typ-
considered a means to manipulate the cause “neuronal ically less thoroughly elaborated than others (e.g., the kind
activity” (via an E-field), whereas behavioral measures are of neuronal activity induced by the E-field or the precise
used as an observable proxy to assess the hidden effect behavioral changes to be expected from changes in cogni-
“cognitive function.” Every single link in this causal path tive function), crucial assumptions about mediators re-
can be confounded by several (known or unknown) factors main untested (e.g., whether the TMS-induced E-field
(yellow, blue, and green boxes in Figure 1B), which are actually produced the neuronal activity that was aimed
best experimentally eliminated or controlled for to rule for), and potential confounds are uncontrolled (e.g., the
out alternative explanations for the observed data and to sensory input because of peripheral costimulation).
draw strong conclusions regarding the hypothesized
cause–effect relationship. This is complicated by the facts
GENERAL APPROACHES FOR NIBS IN
that (i) many of the mediating and confounding variables
COGNITIVE NEUROSCIENCE
are not directly observable and have to be approximated
by imulation (E-field), neuroimaging (neuronal activity), Although NIBS is often depicted as a means by which we
or modeling (cognitive function) and (ii) few of the rele- can simply “switch off” or “knock out” entire brain regions
vant cause–effect pairs express linear dose–response or realistically mimic endogenous oscillatory activity, the
Figure 2. Schematic representation of noninvasive brain stimulation approaches. (A) Online quantification: a stimulation strong enough to cause a
direct output of the targeted region/network (with TMS, not TCS) that allows to quantify cortical excitability via MEPs or phosphene reports. (B)
Online interference: a disruption of ongoing task-related or spontaneous brain activity (with TMS, rather not TCS) that disturbs a cognitive function.
(C) Online modulation: a moderate modulation of the level (“gating” via low-intensity TMS or TDCS) or timing (“entrainment” via TACS or rhythmic
TMS) of neuronal activity that interacts with ongoing task-related or spontaneous neuronal activity without disrupting it. (D) Offline facilitation: an
increase in cortical excitability (triggered by rTMS or prolonged TCS) presumably mediated via LTP of the stimulated synapses. (E) Offline inhibition:
a decrease in cortical excitability (triggered by rTMS or prolonged TCS) presumably mediated via LTD of the stimulated synapses.